|Publication number||US6600805 B2|
|Application number||US 10/117,237|
|Publication date||Jul 29, 2003|
|Filing date||Apr 8, 2002|
|Priority date||Oct 21, 1999|
|Also published as||CA2387756A1, CA2387756C, DE60010499D1, DE60010499T2, EP1226431A2, EP1226431B1, US20020168046, WO2001029557A2, WO2001029557A3|
|Publication number||10117237, 117237, US 6600805 B2, US 6600805B2, US-B2-6600805, US6600805 B2, US6600805B2|
|Inventors||Per Waaben Hansen|
|Original Assignee||Foss Electric A/S|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (21), Non-Patent Citations (6), Referenced by (35), Classifications (14), Legal Events (5)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is a continuation of PCT International Application No. PCT/DK00/00588 filed Oct. 20, 2000.
The present invention relates to X-ray analysis, and more specifically to the determination of properties of food or feed, such as the fat content of meat.
X-ray analysis for determining the fat content of meat has been known for several years. Such examples are described in numerous documents. U.S. Pat. No. 4,168,431 (Henriksen) discloses a multiple-level X-ray analysis for determining fat percentage. The apparatus includes at least three X-ray beams at different energy levels. DK PS 172 377 B1 discloses detection means for X-rays as well as a system for determination of properties of an item by use of X-rays. The system operates at a single energy level and applies two detection means separated by a X-ray attenuating material.
WO 92/05703 discloses a method and device for cutting food products. The positioning of suitable cuts are guided by use of X-ray scanning showing the distribution of tissue type in the product. U.S. Pat. No. 5,585,603 discloses a method and system for weighing objects using X-rays. A continuous X-ray analysis for a meat blending system is known from U.S. Pat. No. 4,171,164 (Groves et al).
The percentages of fat in two meat streams are determined by passing a beam of polychromatic X-rays through the streams, measuring both the incident and the attenuated beams. U.S. Pat. No. 4,504,963 discloses an apparatus, system and method for determining the percentage of fat in a meat sample through use of X-ray radiation techniques. An automatic calibration is obtained by use of three incident beams, all at same energy level. Validation of body composition by dual energy X-ray absorptiometry is described in Clinical Physiology (1991) 11, 331-341. (J. Haarbo, A. Gotfredsen, C. Hassager and C. Christiansen). Further studies on bodies are reported in Am. J. Clinical Nutrition 1993: 57:605-608. (Ole Lander Svendsen, Jens Haarbo, Christian Hassager, and Claus Christiansen).
Recently analysis of meat has been reported in Meat Science Vol. 47, No 1/2, 115-124, 1997 (A. D. Mitchell, M. B. Solomon & T. S. Rumsey). A thorough study on pork carcasses by use of dual-energy X-ray absorptiometry was reported by P Elowsson et al (1998) J. Nutr. 128 1543-1549 An Evaluation of Dual-Energy X-Ray Absorptiometry and Underwater Weighing to estimate Body Composition by means of carcass Analysis in Piglets. (p. 1543, l.&r. col.; p. 1544, l.col.; p. 1547, l. col.). Another analysis on pork carcasses is reported by Mitchell et al., J. Anim. Sci. (1998), vol. 76, pp 2104-2113. However, on page 2113 of this analysis it is specifically concluded that the X-ray analysis is too slow for compatibility with on-line processing. None of the above-mentioned prior art has so far lead to an efficient apparatus fulfilling the needs in a slaughterhouse. Generally the prior art shows difficulties when measuring layers of varying thickness, specifically thin layers adjacent to thick layers. Further the prior art is unable to measure and provide results as fast as required to be useful for online processing.
The presently applied apparatus in most slaughterhouses is a Continuous Fat Analyser (Wolfking A/S, Denmark) and Infratec 1265 (Foss Tecator AB, Sweden) using NIR technology. Also applied is Anyl-Ray (The Kartridg Pak Co., Iowa) using a single energy X-ray on a sample of well-defined weight or volume.
It is an object of the present invention to provide a method and apparatus enabling a faster and more accurate determination than hitherto known, of the fat content in a food or feed product, such as a batch of meat trimmings, allowing creation of specific products (such as sausages or minced meat) having a desired content of fat, which is much more accurate than presently possible.
The present invention also applies regression analysis and multivariate calibration. Such analysis is known from e.g. the applicant's own WO 95/16201 disclosing the Determination of extraneous water in milk samples using regression analysis and multivariate calibration. Further, the applicant's WO 98/43070 discloses Measurement of acetone in milk using IR spectroscopy and multivariate calibration. U.S. Pat. No. 5,459,677 discloses a calibration transfer for analytical instruments. The applicant's WO 93/06460 discloses an infrared attenuation measuring system, including data processing based on multivariate calibration techniques, and the applicant's U.S. Pat. No. 5,252,829 discloses a determination of urea in milk with improved accuracy using at least part of an infrared spectrum.
The present invention relates to a method of determining properties a medium of food or feed, such as the fat content of meat, by use of dual X-ray absorptiometry, the medium being a raw material of food or feed, a product or intermediary product of food or feed, or a batch, sample or section of the same, the method comprising—scanning substantially all of the medium by X-ray beams having at least two energy levels, including a low level and a high level,—detecting the X-ray beams having passed through the medium for a plurality of areas (pixels) of the medium,—for each area calculating a value, Alow, representing the absorbance in the area of the medium at the low energy level,—for each area calculating a value, Ahigh representing the absorbance in the area of the medium at the high energy level, characterised by for each area generating a plurality of values being products of the type Alow n*Ahigh m wherein n and m are positive and/or negative integers or zero, and predicting the properties of the medium in this area by applying a calibration model to the plurality of values, wherein the calibration model defines relations between the plurality of values and properties of the medium.
The advantage over the prior art is a more accurate determination of the properties, such as the fat content in the medium. The accuracy is specifically improved over the prior art when measuring layers of varying thickness. A further advantage is due to the fact that using the method according to the invention almost the whole product is measured instead of a sampling. Generally, when using sampling in an inhomogeneous medium the extraction of a sample will introduce an error, because the sample may not be representative.
Preferably the plurality of values includes values Alow n1/Ahigh m1, wherein n1 and m1 are positive integers. Further on it is preferred that the plurality of values includes the values Alow, Ahigh, Alow 2, Ahigh 2, and Alow/Ahigh and/or at least one of the values Alow*Ahigh; Alow 2*Ahigh; Alow*Ahigh 2 and/or at least one of the values Alow*Ahigh; Alow 2*Ahigh; Alow*Ahigh 2; Alow*Ahigh 4 and Alow 2*Ahigh 4 and/or at least one of the values Alow 2/Ahigh; Alow/Ahigh 2 and Alow 2/Ahigh 2; Alow 3/Ahigh 2; Alow 4/Ahigh 2; 1/Ahigh 4; Alow 4/Ahigh 3 ; A low 3/Ahigh4 and Alow 4/Ahigh 4.
Practical experiments have proved that such values contribute considerably to improve the accuracy.
Preferably the calibration model is obtained by use of a regression method being included in the group comprising Principal Component Regression (PCR), Multiple Linear Regression (MLR), Partial Least Squares (PLS) regression, and Artificial Neural Networks (ANN).
The present invention further relates to an apparatus for the determination of properties of a medium, such as the content of a component in the medium, the medium comprising a raw material of food or feed, a product or intermediary product of food or feed, or a batch, sample or section of the same, the apparatus comprising means (12, 14) for emitting at least two X-ray beams (16, 18) at two different energy levels, means for directing the at least two X-ray beams towards and through the medium, X-ray detection means (22, 24) covering a plurality of areas for detecting the two beams (16, 18) after passing through the medium, means (27, 28, 34, 35) for transferring and converting output signals from the detection means (22, 24) into digital data set for input to data processing means (38) for receiving, storing and processing the at least two data set representing X-ray images at the at least two different energy levels, the apparatus further comprising means for synchronising the at least two data sets and the data processing means including means for calculating values representing the absorbances (Alow, Ahigh) in each area of the medium at the at least two energy levels, characterised in that the data processing means comprise means for generating a plurality of values being products of the type Alow n*Ahigh m wherein n and m are positive and/or negative integers or zero, and means for predicting the properties of the medium in this area by applying a calibration model to the plurality of values, wherein the calibration model defines relations between the plurality of values and properties of the medium.
The advantage over the prior art is a faster and more accurate determination, which is so fast that it can be applied continuously on a process line in a slaughterhouse. The medium is arranged on a conveyor moving at substantially constant speed, and the at least two X-ray beams are fan-shaped, and the low level beam is detected by a first linear array, being dedicated to the detection of the low energy beam, and the high level beam is detected by a second linear array being dedicated to the detection of the high energy beam, each comprising a plurality of pixels.
The apparatus may be characterised by comprising at least one low energy X-ray source (12) arranged above the medium (20) for providing a fan-shaped low energy beam (16) substantially covering the width of medium and at least one high energy X-ray source (14) arranged above the medium (20) for providing a fan-shaped low energy beam (16) covering the width of medium (20) and a first X-ray detection means (22) arranged to be exposed to the fan-shaped low energy beam (16) and below the medium (20) a second X-ray detection means (24) arranged to be exposed to the fan-shaped high energy beam (18) and below the medium (20), and electronic means (34, 38, 42) including the data processing means (38) and communicating with the detectors (22, 24) and arranged to store and process data representing signals from the detection means (22, 24), and further comprising means (10) for moving the medium (20) relative to the X-ray beams (16, 18) or visa versa.
The apparatus may be characterised in that the data processing means include and/or communicate with means including data storage means comprising a calibration model prepared by use of multivariate calibration methods such as Artificial Neural Networks (ANN), or PCR, MLR or PLS regression analysis.
The apparatus may be characterised by comprising at least two sources (12, 14) emitting X-rays of two different energy levels.
The apparatus may be characterised by the two energy levels comprising a low energy level in a range between 35 and 75 keV, preferably between 45 and 70 key and most preferred about 62 key, and a high energy level in a range between about 60 and 140 keV, preferably between 80 and 130 key and most preferred about 120 keV.
The apparatus may be characterised by comprising filter means located in each of the beams (16, 18)
The apparatus may be characterised by comprising one X-ray source and two filter means splitting the beam into two beams of X-rays at two different energy levels.
The apparatus may be characterised in that the means (12, 14) for emitting at least two X-ray beams, the means for directing the at least two X-ray beams and the X-ray detection means (22, 24) are mutually fixed.
The apparatus may be characterised by comprising means (12, 14) for emitting spatially separated fan-shaped beams (16, 18).
The apparatus may be characterised in that the detection means (22,24) are covered by a scintillating layer, e.g. cadmium telluride, mercury iodide, and/or gadolinium oxysulphide.
The apparatus may be characterised by comprising conveyor means (10) arranged to carry container means (20), such as a tray or an open box, adapted to accommodate a random number of meat lumps of various sizes to be analysed, the conveyor means being arranged to let the container means (20) pass the at least two fan-shaped X-ray beams (16, 18).
The apparatus may be characterised by comprising conveyor means (10) wherein the conveyor belt is made from a material showing a low absorption of X-rays, and/or is split into two separate, spaced parts, the detector means (22, 24) being arranged in an open space between the two parts.
The apparatus may be characterised by comprising conveyor means (10) adapted to accommodate a Continuous flow of meat lumps of various sizes to be analysed, the conveyor means being arranged to let the meat lumps pass the at least two fan-shaped X-ray beams (16, 18).
The apparatus may be characterised by being arranged to perform the following steps: scan at least a section of a medium by X-ray beams having at least two energy levels, store data representing at least two X ray images of the medium, calculate the fat content and/or areal density for all points (pixels) obtained from the scanning by use of multivariate calibration models generated in a previously performed calibration step, multiply the fat content and areal density at each point, in order to generate a “fat map” (in g/cm2) of the sample, add all points in the “fat map” to give the total fat weight (Ftotal) of the sample, add all areal densities for the sample to give the total weight (Wtotal) of the sample, calculate the average fat content of the sample as the ratio Ftotal/Wtotal.
The present invention further relates to a method for calibration of an apparatus comprising preparation of a plurality of calibration samples consisting of specified food or feed products, such as minced pork meat, of various well-defined heights and properties, measuring the plurality of calibration samples in the apparatus, thereby obtaining data representing two X-ray responses of each sample, each response comprising a plurality of pixels, and wherein the data of each pixel, or the mean of a number of neighbouring pixels, are processed using the formulas:
or similar expressions for calculation of values representing the absorbance in an area of the medium above a pixel or a number of neighbouring pixels,
generating a plurality of values of the type Alow n*Ahigh m,, wherein n and m are positive and/or negative integers and/or zero, correlating—by use of multivariate calibration methods, such as Artificial Neural Networks (ANN), or PCR, MLR or PLS regression—the data set for all/or a plurality of calibration samples to the properties determined by other means, such as a reference method,—in order to determine a number of calibration coefficients, providing a calibration model comprising the number of determined calibration coefficients.
All calibration samples are prepared in such a manner that they are homogeneous and of fixed areal densities, and further by averaging each of the values over all pixels at least in a defined portion of the images.
The invention further relates to a method of predicting the fat content of meat, comprising use of a calibration model obtained by a method. The invention also relates to an apparatus comprising a calibration model determined by a particular method.
By use of the present invention it is possible—more accurate and more rapidly than hitherto known—to determine the fat content of a random number of meat lumps (such as trimmings or cuts) of various sizes in a container (or similar means for enclosing or carrying a load of meat) or directly on a conveyor belt. The measurement may be performed within a fairly short time, such as a few seconds, e.g. about 4.5 or 9 seconds per container, each container having a volume of e.g. about 0.1 m3. Preferably, a smaller volume, about e.g. 25 kg meat, is arranged in each container. Accordingly, the method and apparatus may be applied for on-line control of the production of various meat products, such as minced meat, and more specifically where minced meat is produced from meat trimmings of various sizes.
According to the applicant's best knowledge multivariate calibration techniques have never been applied to X-ray analysis of meat, nor to X-ray analysis in general. The use of multivariate techniques solves a specific problem present when using the techniques according to the prior art. The known apparatus becomes highly inaccurate when measuring on a combination of thin and thick layers. When measuring meat lumps of various sizes the thickness of the layers through which the X-ray has to pass will vary considerably from 0 or almost 0 to a specified maximum. The use of a plurality of values allows a better accuracy of such measurements than hitherto known.
Further scope of the applicability of the present invention will become apparent from the detailed description given hereinafter. However, it should be understood that the detailed description and specific examples, while indicating preferred embodiments of the invention, are given by way of illustration only, since various changes and modifications within the spirit and scope of the invention will become apparent to those skilled in the art from this detailed description.
The present invention will become more fully understood from the detailed description given hereinbelow and the accompanying drawings which are given by way of illustration only, and thus are not limitative of the present invention, and wherein:
FIG. 1 shows as an example a system according to the invention
FIG. 2 shows a preferred embodiment of an X-ray apparatus according to the invention
FIG. 3 shows a simulation of a system comprising one source and a combination of two filters.
FIG. 4 shows cross-validated X-ray fat predictions versus reference fat content of 32 calibration samples when performing a simple univariate regression of Alow/Ahigh against the reference fat content of the samples.
FIG. 5 shows cross-validated X-ray fat predictions versus reference fat content of 32 calibration samples when performing a PLS calibration with 5 PLS factors (based on 11 variables) against the reference fat content of the samples.
FIG. 6 shows cross-validated X-ray areal densities versus reference areal density of 32 calibration samples when performing a simple univariate regression of Ahigh against the reference areal densities of the samples.
FIG. 7 shows cross-validated X-ray areal densities versus reference areal density of 32 calibration samples when performing a PLS calibration with 1 PLS factor (based on 2 variables) against the reference areal densities of the samples.
FIG. 8 shows Fat predicted by X-ray in 99 points of a meat sample.
FIG. 9 shows Areal density predicted by X-ray in 99 points of a meat sample.
FIG. 10 shows Fat (in g/cm2) predicted by multiplication of the fat content (FIG. 8) by the areal density (FIG. 9) in 99 points of a meat sample.
FIG. 11 shows a flow diagram illustrating the measuring process.
FIG. 12 shows a typical meat sample in a plastic container.
FIG. 13 shows a typical low energy X-ray transmission image of a meat sample as shown in FIG. 12.
FIG. 14 shows a typical high energy X-ray transmission image of the same meat sample.
FIG. 15 is an image illustrating a calculated areal density for each individual pixel.
FIG. 16 is an image illustrating a calculated fat content for each individual pixel.
FIG. 17 is an image illustrating a calculated “fat map” for a meat sample of 36% fat.
FIG. 18 shows a reference versus predicted plot for 50 scans.
The following description discloses as an example a preferred embodiment of the invention using two X-ray sources. The apparatus is designed for being installed in relation to a production line in a slaughterhouse. FIG. 1 shows a schematic diagram of an embodiment of a measurement system according to the invention. FIG. 2 illustrates the principle of the presently preferred X-ray apparatus. FIG. 2 shows only the active operating portions of the X-ray equipment. For purpose of clarity, all protective shielding or screening and all casings are deleted from the drawing. The equipment comprises or is located in close relation to a conveyor 10. Two X-ray sources 12, 14 are arranged above the conveyor 10. From the two sources 12, 14 X-ray beams 16, 18 are directed towards detectors 22, 24 arranged below the conveyor. The conveyor may be split into two separate conveyors spaced to allow free pass of the X-rays and to leave an open space for location of detectors 22, 24. Alternatively the conveyor belt should be made from a material showing a low absorbance of X-rays, e.g. polyurethane or polypropylene. The food or feed to be measured is arranged in an open container or box 20, preferably also composed by a material showing low absorbance of X-rays. Obviously in an alternative arrangement the sources could be located below the conveyor and the detectors above the conveyor.
The operational speed of the conveyor is preferably substantially constant. The items, motor 30, control box 33, and cables 32, 39, shown by phantom lines in FIG. 1, indicate that the operation of the conveyor optionally may be controlled by the computing means 38. The conveyor may include position measuring means, e.g. an encoder installed on a conveyor driving shaft. Alternative means may be a laser or radar detection or marks on the conveyor belt. It is essential to the present method that the data representing the two X-ray images can be synchronised. Such synchronisation may however be obtained in many ways, including mathematical post-processing of the images.
The equipment used in the present example consists of two constant potential X-ray sources 12, 14, one at low energy (e.g. 62 kV/5 mA) and another at high energy (e.g. 120 kV/3 mA), both with an appropriate filtration (e.g. using 0.25 and 1.75 mm of copper, respectively) narrowing the spectral range of the radiation emitted from the polychromatic sources. The two sources are spatially separated to avoid interference between them, i.e. to avoid that radiation from one source is detected as if it originated from the other. The radiation from either source is collimated by a lead collimator. In this way two fan-shaped beams of X-rays 16, 18 are directed through container 20 comprising a sample or batch of the food or feed product towards detectors 22, 24, e.g. Hamamatsu C 7390. Alternatively the meat lumps may be arranged loosely on a conveyor band.
Further, the two separate sources may be replaced by a combination of one source and two filters emitting a low energy and a high energy beam. The resulting source spectra are shown in FIG. 3. However the preferred embodiment applies two separate sources 12, 14 driven by separate power supplies 13, 37.
Both X-ray sources 12, 14 are associated with an array of detectors 22, 24 covered with a scintillating layer converting the transmitted radiation into visible light that can be measured by the detectors 22, 24. The scintillating layer may consist of e.g. cadmium telluride, mercury iodide, and/or gadolinium oxysulphide. The pixels used in the presently preferred embodiment have the dimensions 1.6×1.3 mm2 and are arranged as an array of 384 pixels with a pitch of 1.6 mm. These dimensions are only stated as an example. Other dimensions may be applied. The pixels convert the amount of transmitted light into analogue signals that are passed through cables 27, 28 to an analogue-to-digital converter 34 which is connected through cable 35 to a computing means 38 capable of performing the successive calculations. A monitor 42 may be connected through cable 40 to the computing means to show results or details of the operation. The computing means 38 may include means for controlling the supply of power through means 36, 37, 26 and 25, 13, 15 to the X ray sources 12, 14. The monitor 42 and the computing means 38 may comprise a Personal Computer, preferably including at least one Pentium processor and/or a number of digital signal processors.
A container 20, comprising e.g. meat trimmings from a cutting section of the slaughterhouse, is received on the conveyor 10. The container is moved with a fairly constant speed of e.g. about 5-100 cm per second, such as 10-50 cm, e.g. 30 cm per second past the fan shaped beams 16, 18 and the arrays of detectors 22, 24 in a controlled manner in order to generate two images of the sample or batch, one at a low X-ray energy and another at a high energy. All data representing the two images are stored in the computer 38.
FIG. 11 represents a flow chart illustrating the measurement and data treatment. As stated above, two X-ray images of each container, comprising a batch of food or feed e.g. meat, are obtained. The signals at the pixels are Ilow and Ihigh at low and high X-ray energies, respectively, (110, 112 in FIG. 11). Furthermore, the so-called “dark signals” (i.e. the signal from the detectors when no radiation reaches them), Idark(low) and Idark(high), and the “air signals” (i.e. the signal from the detectors when no sample is present in the sampling region), Iair(low) and Iair(high), are collected for each pixel at both X-ray energies (102 in FIG. 11). Preferably these data are collected repetitively in the intervals between the passage/passing of meat containers, i.e. the dark signals and air signals are measured repetitively, e.g. at regular intervals during a day to adjust for any drift of instrument performance.
Now referring to 114 in FIG. 11, these signals are transformed into absorbance units by using the following formulas:
From these two values, a plurality of values can be generated e.g.: Alow; Ahigh; Alow 2; Ahigh 2; Alow×Ahigh; Alow 2×Ahigh; Alow×Ahigh 2; Alow/Ahigh; Alow 2/Ahigh; Alow/Ahigh 2; (Alow/Ahigh)2;
or in a more generalised manner: Alow n*Ahigh m,
wherein n and m are positive and/or negative integers and/or zero,
These values are used as the input for the calibration routine establishing a relationship between the collected data and the component (e.g. the fat content) or the property (e.g. the areal density) of interest.
It is essential that a value Alow for a specific pixel measuring the low energy transmittance through a specific area of the medium is matched to, the value Ahigh for the pixel measuring the high energy transmission through exactly the same area of the medium. This can be accomplished by ensuring a synchronisation of the pictures as mentioned below.
If the low and high energy images are not perfectly aligned, i.e. if a specific region of the sample does not show up at exactly the same positions in the two images, large errors may result. This problem may occur e.g. if the two line scan detectors (22, 24) are not synchronised. A possible solution to this problem is to calculate the correlation between the two images using various shifts between them and thereby finding the shift at which the correlation is at a maximum, followed by a correction of one of the images by this shift. It is however preferred to synchronise the line scanning e.g. by the use of a position measuring means, or by tight control of the conveyor speed.
The following example explains how to generate a calibration model.
A set of 32 calibration samples consisting of minced pork meat were prepared. They were frozen in blocks of varying heights (5, 10, 15, and 20 cm) with horizontal dimensions of 10×10 cm2. Their fat content (percentage) which ranged from 2.6 to 70.9%, were later determined by use of a wet chemistry method. The heights and fat contents (percentage), together with the fat-dependent density of meat, were used for calculating the areal densities of all 32 samples, ranging from 4.8 to 21.0 g/cm2.
The frozen meat blocks were measured in the aforementioned X-ray equipment, yielding two images of each sample. The data points (pixels) of these images were treated according to the steps described above. To avoid random noise from influencing the calibration results, the 11 values generated from the original absorbance values were averaged over all pixels in the image. This could only be done since the samples were homogeneous and of fixed height.
This data set consisting of 11 variables obtained for all 32 samples was correlated against the fat content (percentage) measured by a reference method and the areal densities using the Partial Least Squares (PLS) regression method. This, and other similar multivariate calibration methods are well known (Martens and Næs: Multivariate Calibration, 2nd ed., Wiley (1992)).
The calibrations were validated using full cross-validation, i.e. one sample at a time was removed from the data set for validation while the remaining 31 samples were used for calibration. This procedure was repeated for all samples, and validation results were generated by combining the validation results for all 32 samples.
The traditional way of building an X-ray calibration model for fat in meat is by correlating the Alow/Ahigh ratio to the fat reference results (Haardbo et al., Clin. Phys. (1991), vol. 11, pp. 331-341 or Mitchell et al. J. Anim. Sci. (1998), vol. 76, pp. 2104-2114). This method is, however, sensitive to the thickness (or areal density) of the sample and is therefore not useful with the range of sample heights (from 5 to 20 cm) of interest in the present context. This is evident from FIG. 4, where X-ray fat predictions using only the Alow/Ahigh ratio are plotted against the fat reference results. The prediction error (expressed as the Root Mean Square Error of Prediction, RMSEP) is 14.7% in this case.
Using the method according to the invention, with e.g. 11 or more variables generated from the original two absorbences in combination with a PLS regression with 5 PLS-factors, the plot presented in FIG. 5 is obtained. In this case, the prediction error (RMSEP) is as low as 1.0%, thus showing the benefits of using the PLS method in combination with the new variables.
The method can also be applied for the determination of the areal density of the sample. According to the prior art the areal density is determined by correlating Ahigh to the reference areal density. The result of such a calibration model is presented in FIG. 6, where the agreement between the areal density determined by X-ray and the reference results is very good. The prediction error (RMSEP) is 0.30 g/cm2 in this case. When using the method according to the invention, i.e. using both measured absorbences, Alowand Ahigh, in combination with a PLS regression with 1 factor, the result presented in FIG. 7, and a prediction error (RMSEP) of 0.28 g/cm2, is obtained. This is only a slight improvement, but the use of two variables instead of one provides the user with a further advantage: the possibility of detecting incorrect measurements (e.g. if one of the two X-ray sources shows a sudden drop in intensity, or if a pixel is not responding). This is because discrepancies from the relationship between Alow and Ahigh can easily be detected by the PLS model. Such outlier detection is not possible if only one absorbance is used. This possibility is very relevant and advantageous when using CCD detector wherein a single pixel may deteriorate fairly abruptly.
The calibration models developed in this way can be used for future predictions of the fat content and areal density in a given point in an inhomogeneous meat sample as well as for determination of the mean fat content of a large meat sample.
The following example will demonstrate the use of the calibration models in practice where samples are inhomogeneous and of varying thickness. The purpose is to predict the mean fat content of the samples. Therefore the procedure involves the following steps as shown in FIG. 11:
1. Regular measuring of Idark and Iair, 102
2. Arranging a batch or stream of meat (or other food or feed product) on a conveyor passing through the X-ray apparatus, 104
3. Scanning the batch or stream by X-ray beams at two different energy levels, 106, 108,
4. Detecting signals representing a plurality of X-ray intensities, using the detectors 22, 24 in FIGS. 1, 2 110, 112
5. Recording data representing the detected signals 114.
6. Calculate Alow and Ahigh for all pixels 114, (optionally, a smoothing of the picture may be included.)
7. Co-ordinate (match) Alow values and Ahigh values 114, if necessary.
8. Calculate derived expressions Alow n*Ahigh m 114.
9. Calculate the fat content (percentage) and preferably the areal density for all points (pixels) obtained from the scannings, using a fat calibration model generated as described above 116.
10. Multiply the fat content (percentage) and areal density at each point, in order to generate a “fat map” (in g/cm2) of the batch or stream of food or feed 116.
11. Add all points in the “fat map” to give the total fat weight (Ftotal) 116.
12. Add all areal densities for the sample to give the total weight (Wtotal) 116.
13. Calculate the average fat content (percentage) as the ratio Ftotal/Wtotal 116.
Optionally, two more steps may be included between step 6 and 7:
If the meat lumps are arranged in a container the data should be subjected to a correction for the absorption in the bottom of the container. Such correction is preferably made at the end of step 6, providing new corrected values of Alow and Ahigh for all pixels.
A further advantageous option is smoothing of the data, e.g. in the direction of the movement.
Further experience has proved that it can be advantageous to include a further data processing in step 9. In a presently preferred embodiment pixels having an areal density outside a specified interval are removed/deleted or at least disregarded in the following data processing, i.e. pixels for which the calculated areal density is extremely low or much too high, are rejected.
The present example shows the calculation of the fat content (percentage) for one row of 99 points only. This is done in order to make the presented plots simpler, and is easily generalised to be performed on a two-dimensional image of a meat product.
A meat sample consisting of a cubic block (dimensions: 10×10×10 cm3) of minced pork meat was measured using the same X-ray equipment as was used for measuring the calibration samples. The first two steps of the prediction procedure are shown in FIG. 8 (predicted fat content (percentage) ), FIG. 9 (predicted areal density, step 9), and FIG. 10 (the “fat map”, step 10). There is clearly a variation in both fat content and areal density over the sample, so the results from all sample points are needed in order to obtain an accurate estimate of the mean fat content of the sample.
The sum of all points in the “fat map” (step 11), Ftotal, equals 464 g/(99 pixels), and the total weight of the sample (step 12), Wtotal, is 963 g/(99 pixels). This, in turn, results in a predicted mean fat content of 464/963=48.2% (step 13) , not far from the true fat content, which was determined later as 49.2% by a reference method.
This example is an extension of the results stated above. The present example involves the prediction of the fat content of samples consisting of approximately 25 kg of meat in plastic containers.
Ten samples of meat ranging from 11.5 to 84.6% fat were obtained from a meat processing plant. The amount of meat in each container ranged from 20 to 30 kg, the sample homogeneity ranging from ground meat to meat pieces of 5 kg each. A typical sample consisting of 36% fat trimmings arranged in a presently preferred container (dimensions: 70×40×17 cm3) is shown in FIG. 12.
Each of these ten samples were scanned by the instrument five times over a period of two days. Before each new scan the contents of the container was reorganised, i.e. the meat pieces were moved around without changing the total content of the container. This was done in order to check the repeatability of the measurement. A total of 50 X-ray scans, each consisting of a low and a high energy image of 306×1836 data points, were thus gathered. Two typical transmission images of a sample are shown in FIGS. 13 and 14.
All 50 scans were subjected to the prediction steps according to the invention, using a calibration model based on frozen meat samples. The calculated fat content and areal density for each individual pixel (step 9), as well as the “fat map” (step 10) for one sample are shown in FIGS. 16, 15, 17. The negative fat predictions are due to a relatively low signal-to-noise ratio on the individual pixels. This is,however, no problem, as the final averaging of the results reduces this error by orders of magnitude. From these images, the total fat content of the samples was calculated. The pooled repeatability standard deviation, sr, for the five different scans of the same sample was 0.25%.
After this experiment had been carried out the samples were homogenised and a number of subsamples were analysed by the reference method for fat in meat (Schmid-Bondzynski-Ratzlaff, SBR method). These reference results were compared to the predictions obtained above, resulting in an accuracy (Root Mean Square Error of Prediction, RMSEP) of 0.81%. The reference versus predicted plot for the 50 scans is shown in FIG. 18.
To demonstrate the advantage of the method, in terms of its ability to significantly improve the accuracy of the fat determination, a further experiment was carried out. 45 frozen meat samples with fat contents ranging from 2.4 to 72.8% and areal densities ranging from 1 to 21 g/cm2 were measured using the X-ray equipment. These samples were used for obtaining six different calibration models, using various combinations of the 11 variables based on Alow and Ahigh described above. Subsequently, these calibration models were tested on the same data set as used in Example 2, i.e. ten meat samples of 20 to 30 kg with fat contents ranging from 11.5 to 84.6% fat.
The six combinations of variables used for calibration models are shown in the table presented below, along with the resulting accuracies (RMSEP) on the calibration set (cross-validated) and the test set. Furthermore, the repeatability (sr) on the test set was also calculated.
Alow × Ahigh
Alow 2 × Ahigh
Alow × Ahigh 2
Alow/ Ahigh 2
From the results presented in the table the accuracy obtained when using only powers of Alow and Ahigh (calibration model 1) as well as products thereof (calibration model 2) are unacceptable if the method is to be used for process control within tight limits. If the Alow/Ahigh ratio is added, combined with powers of Alow and Ahigh (calibration model 3), an acceptable accuracy is obtained. However, if powers of Alow and Ahigh are combined with Alow/Ahigh and the more complex ratios (calibration model 4), even more accurate predictions result. It is also clear from calibration models 5 and 6 that of Alow and Ahigh and powers thereof are essential if the best possible accuracy is required.
In terms of the repeatability it is also clear that the Alow/Ahigh ratio has a major influence on the difference between multiple determinations of the same sample.
The example presented above demonstrates the advantages in using higher order ratios for calibration of X-ray data against fat reference results. Only orders up to two were used in the present example, but ratios of higher orders may improve the result even further. For example, when using the ratios: (Alow/Ahigh)3; (Alow/Ahigh)4; Alow 3/Ahigh; Alow 4/Ahigh; Alow 3/Ahigh 2; Alow 4/Ahigh 2; Alow 4/Ahigh 3, an accuracy (RMSEP) of 0.67 is obtained on the calibration set.
The method may be applied to all kinds of meat, such as beef, veal, pork, buffalo, camel and lamb, game, such as rabbit, poultry, such as chicken, turkey, duck, goose and ostrich, and fish.
While a single particular embodiment of the invention has been mentioned, it will be understood, of course, that the invention is not limited thereto since many modifications may be made, such as using more than two X-ray sources or alternative arrangements such as arranging the sources below the conveyor or sidewards, and it is, therefore, contemplated to cover by the appended claims any such modifications as fall within the true spirit and scope of the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4168431 *||Jan 6, 1978||Sep 18, 1979||The Kartridg Pak Co.||Multiple-level X-ray analysis for determining fat percentage|
|US4171164||Dec 12, 1977||Oct 16, 1979||The Kartridg Pak Co.||Continuous X-ray analysis for meat blending system|
|US4504963 *||Apr 14, 1983||Mar 12, 1985||Johnson Lloyd D||Automatically calibrating meat analysis device|
|US5252829||May 19, 1992||Oct 12, 1993||A/S Foss Electric||Method of determining urea in milk|
|US5428657||Mar 22, 1994||Jun 27, 1995||Georgia Tech Research Corporation||X-ray monitoring system|
|US5459677||Jun 24, 1991||Oct 17, 1995||Board Of Regents Of The University Of Washington||Calibration transfer for analytical instruments|
|US5585603||Dec 23, 1993||Dec 17, 1996||Design Systems, Inc.||Method and system for weighing objects using X-rays|
|US5682411||May 18, 1994||Oct 28, 1997||St. John Innovation Centre||Imaging system|
|US5841833 *||Mar 11, 1997||Nov 24, 1998||Lunar Corporation||Dual-energy x-ray detector providing spatial and temporal interpolation|
|US6173038 *||Dec 1, 1999||Jan 9, 2001||Cyberlogic, Inc.||Plain x-ray bone densitometry apparatus and method|
|US6233473 *||Apr 15, 1999||May 15, 2001||Hologic, Inc.||Determining body composition using fan beam dual-energy x-ray absorptiometry|
|US6285740 *||Oct 13, 1999||Sep 4, 2001||The United States Of America As Represented By The Secretary Of The Navy||Dual energy x-ray densitometry apparatus and method using single x-ray pulse|
|US6370223 *||Apr 6, 2001||Apr 9, 2002||Ut-Battelle, Llc||Automatic detection of bone fragments in poultry using multi-energy x-rays|
|US6385284 *||Jul 8, 1999||May 7, 2002||Safeline Avs Limited||Radiation monitoring of a physical property of a material|
|DE4406958A1||Mar 3, 1994||Sep 7, 1995||Heimann Systems Gmbh & Co||Scanner für Prüfobjekte|
|DK172377B1||Title not available|
|SE9203711A||Title not available|
|WO1992005703A1||Oct 8, 1991||Apr 16, 1992||Frigoscandia Food Process Systems Ab||Method and device for cutting food products|
|WO1993006460A1||Sep 25, 1992||Apr 1, 1993||A/S Foss Electric||An infrared attenuation measuring system|
|WO1995016201A1||Dec 9, 1994||Jun 15, 1995||Foss Electric A/S||Determination of extraneous water in milk samples, or the freezing point depression of milk samples|
|WO1998043070A1||Mar 20, 1998||Oct 1, 1998||Foss Electric A/S||Measurement of acetone in milk using ir spectroscopy|
|1||*||Bertin, Eric et al., "Evaluation of Dual-Energy Absorptiomettry for Body-Compostion Assessment in Rats," J Nutr., Sep. 1998, vol. 128, Iss 9, pp. 1550-1554.*|
|2||*||Elowsson, Per et al., "An evaluation of Dual-Energy X-ray Absorptiometry and Underwater Weighing to Estimate Body Composition by Means of Carcass Analysis in Piglets," J Nutr., Sep. 1998, vol. 128, Iss. 9, pp. 1543-1549.*|
|3||Haarbo, J. et al., "Validation of body composition by dual energy X-ray absoptiometry (DEXA)", Clinical Physiology (1991), vol. 11, pp. 331-341.|
|4||Mitchell, A.D. et al., "Composition Analysis of Beef Rib Sections by Dual-energy X-ray Absorptiometry", Meat Science (1997), vol. 47, No. 1/2, pp 115-124.|
|5||Mitchell, A.D. et al., "Composition Analysis of Pork Carcasses by Dual-Energy X-Ray Absorptiometry", J. Anim. Sci. (1998), vol. 76, pp. 2104-2114.|
|6||Svendsen, O.L. et al., "Accuracy of measurements of body composition by dual-energy x-ray absorptiometry in vivo", Am. J. Clin. Nutr. (1993), vol. 57, pp. 605-608.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7060981||Sep 7, 2004||Jun 13, 2006||Facet Technology Corp.||System for automated detection of embedded objects|
|US7106830 *||Jun 12, 2002||Sep 12, 2006||Agilent Technologies, Inc.||3D x-ray system adapted for high speed scanning of large articles|
|US7123685 *||Nov 20, 2001||Oct 17, 2006||Convenience Food Systems Wallau Gmbh & Co. Kg||Device and method for analyzing fat of meat|
|US7342221 *||Nov 19, 2004||Mar 11, 2008||Canon Kabushiki Kaisha||Radiation image pick-up device, radiation image pick-up method, and image pick-up system|
|US7386093 *||Nov 23, 2005||Jun 10, 2008||Tsinghua University||Method and an apparatus for liquid safety-detection with a radiation source|
|US7488107||Aug 18, 2005||Feb 10, 2009||General Electric Company||Method and apparatus to detect and correct alignment errors in x-ray systems used to generate 3D volumetric images|
|US7724869 *||Apr 23, 2007||May 25, 2010||Tsinghua University||Detector array and device using the same|
|US8223922 *||Nov 10, 2009||Jul 17, 2012||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, radiation inspection system, and radiation detection method|
|US8280005||Nov 10, 2009||Oct 2, 2012||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, and method for detecting radiation|
|US8306188 *||Apr 26, 2006||Nov 6, 2012||Elisabeth Katz||Method and device for the online determination of the ash content of a substance conveyed on a conveyor, and device for carrying out such an online determination|
|US8483355 *||Sep 10, 2010||Jul 9, 2013||Yokogawa Electric Corporation||Radiation inspection apparatus comprising a gas ejecting unit for supporting and conveying a sheet-like sample|
|US8600005||Aug 31, 2012||Dec 3, 2013||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, and method for detecting radiation|
|US8718230||Jan 21, 2009||May 6, 2014||Pingan Luo||Method and system for determining the constituent content of a multiphase fluid|
|US8964939||Jun 27, 2012||Feb 24, 2015||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, radiation inspection system, and radiation detection method|
|US8981310 *||Jun 29, 2010||Mar 17, 2015||Hamamatsu Photonics K.K.||Radiation detecting device|
|US9329301||Jun 29, 2010||May 3, 2016||Hamamatsu Photonics K. K.||Radiation detecting device|
|US9364106||Mar 2, 2015||Jun 14, 2016||Fitly Inc.||Apparatus and method for identifying, measuring and analyzing food nutritional values and consumer eating behaviors|
|US9594031||Jan 23, 2015||Mar 14, 2017||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, radiation inspection system, and radiation detection method|
|US20030231739 *||Jun 12, 2002||Dec 18, 2003||Rosner S. Jeffrey||3D X-ray system adapted for high speed scanning of large articles|
|US20040081275 *||Nov 20, 2001||Apr 29, 2004||Anders Ostergaard||Device and method for analyzing fat of meat|
|US20050067570 *||Sep 7, 2004||Mar 31, 2005||Retterath James E.||System for automated detection of embedded objects|
|US20050109927 *||Nov 19, 2004||May 26, 2005||Canon Kabushiki Kaisha||Radiation image pick-up device, radiation image pick-up method, and radiation image pick-up system|
|US20060115044 *||Nov 23, 2005||Jun 1, 2006||Wanlong Wu||Method and an apparatus for liquid safety-detection with a radiation source|
|US20070041508 *||Aug 18, 2005||Feb 22, 2007||General Electric Company||Method and apparatus to detect and correct alignment errors in X-ray systems used to generate 3D volumetric images|
|US20070286337 *||Apr 23, 2007||Dec 13, 2007||Xuewu Wang||Detector array and device using the same|
|US20100119038 *||Nov 10, 2009||May 13, 2010||Hamamatsu Photonics K.K.|
|US20100119040 *||Nov 10, 2009||May 13, 2010||Hamamatsu Photonics K.K.||Radiation detection device, radiation image acquiring system, and method for detecting radiation|
|US20100232568 *||Mar 12, 2010||Sep 16, 2010||Heismann Bjoern||X-ray or gamma ray detector array|
|US20100303201 *||Apr 26, 2006||Dec 2, 2010||Elisabeth Katz||Method and Device for the Online Determination of the Ash Content of a Substance Conveyed on a Conveying Means, and Device for Carrying Out Such an Online Determination|
|US20110069813 *||Sep 10, 2010||Mar 24, 2011||Yokogawa Electric Corporation||Radiation inspection apparatus|
|US20120145910 *||Jun 29, 2010||Jun 14, 2012||Hamamatsu Photonics K.K.||Radiation detecting device|
|US20160041110 *||Jul 21, 2015||Feb 11, 2016||Hitachi High-Technologies Corporation||X-ray transmission inspection apparatus and extraneous substance detecting method|
|USD787273||Mar 13, 2015||May 23, 2017||Fitly Inc.||Plate and lid with digital display|
|WO2009135390A1 *||Jan 21, 2009||Nov 12, 2009||Pingan Luo||A method and system for determining the constituent content of a multiphase fluid|
|WO2015003750A1||Jul 11, 2013||Jan 15, 2015||Foss Analytical A/S||Meat processing device incorporating an x-ray analyzer|
|U.S. Classification||378/53, 378/98.9, 702/172, 378/207, 250/252.1|
|International Classification||G01N33/02, G01N33/12, G01N23/06|
|Cooperative Classification||G01N23/06, G01N33/12, G01N33/02|
|European Classification||G01N33/12, G01N23/06, G01N33/02|
|Apr 8, 2002||AS||Assignment|
Owner name: FOSS ELECTRIC A/S, DENMARK
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:HANSEN, PER WAABEN;REEL/FRAME:012769/0581
Effective date: 20011126
|Jul 20, 2004||CC||Certificate of correction|
|Jan 5, 2007||FPAY||Fee payment|
Year of fee payment: 4
|Jan 31, 2011||FPAY||Fee payment|
Year of fee payment: 8
|Jan 29, 2015||FPAY||Fee payment|
Year of fee payment: 12